Method for operating a power device, controller for a power device, and power assembly comprising a power device and such a controller

Abstract

The invention relates to a method for operating a power device (3) using a controller (5). A nominal model, which is made of supply data and which comprises a first mapping between at least one input variable and at least one output variable, and a detail model, which is generated during the operation of the controller (5) using measurement values and which comprises a second mapping between the at least one input variable and the at least one output variable, are used to ascertain at least one deviation variable, which relates to a deviation of the operating behavior of at least one component (6) of the power device (3) from a standard operating behavior, wherein a deviation function is ascertained as the development of the at least one deviation variable over time, and the future development of the at least one component (6) is inferred using the deviation function.

Description

[0001]Rolls-Royce Solutions GmbH DESCRIPTION Method for operating a power device, control device for a power device, and power arrangement with a power device and such a control device. The invention relates to a method for operating a power device, a control device for a power device, and a power arrangement with a power device and such a control device. Deviations in the operation of a power device, such as actuator and / or sensor errors, are typically detected via local trends or limit value monitoring. The disadvantage, however, is that only large deviations from the ideal state can be reliably detected. A more targeted option arises if the power device is controlled in a model-based manner.However, in the case of physical models, this disadvantageously requires a very high level of development effort, which requires identifying suitable models and performing a suitable comparison with measured data. Furthermore, there is an increasing requirement to no longer perform maintenance or component replacement according to rigid schedules, but rather on an as-needed basis and based on actual or predicted wear, defects, or failure probability. The trend statements or forecasts required for this cannot be made easily or with sufficient precision based on local trends or limit value monitoring, whereby the aforementioned disadvantages also apply to physical models.The invention is therefore based on the object of providing a method for operating a power device, a control device for a power device, and a power arrangement comprising a power device and such a control device, wherein the aforementioned disadvantages are at least reduced, preferably not occurring at all. This object is achieved by providing the present technical teaching, in particular the teaching of the independent claims and the preferred embodiments disclosed in the dependent claims and the description.The object is achieved in particular by creating a method for operating a power device, wherein at least one deviation variable is determined based on a nominal model formed from inventory data, which comprises a first mapping between at least one input variable and at least one output variable, and a detailed model created during operation of the control device based on measured values, which comprises a second mapping between the at least one input variable and the at least one output variable, at least one deviation variable is determined, which relates to a deviation of an operating behavior of at least one component of the power device from a standard operating behavior, wherein a deviation function is determined as a temporal development of the at least one deviation variable, and wherein a future development of the at least one component is inferred based on the deviation function.This advantageously allows for a resource-efficient yet reliable forecast for the operation of at least one component, which also enables maintenance or replacement of the at least one component as needed. This approach utilizes the fact that the nominal model, on the one hand, and the detailed model, on the other, provide different mappings between the at least one input variable and the at least one output variable, which can be used to determine the deviation. Since the nominal model is based on existing data, it remains unchanged during ongoing operation of the power device, while the detailed model is expanded based on the measured values ​​during operation.As a result, the detailed model always contains up-to-date information on the actual operating behavior of the power device. When the power device is functioning as intended, in particular when the actuators and sensors are functioning correctly, it is systematically expanded, adapted, and thus improved. However, if a deviation occurs in the operation of the power device, for example an actuator and / or sensor fault, the creation and, in particular, the expansion of the detailed model based on measured values ​​leads to it systematically deviating from the nominal model, resulting in a systematic deviation between the models during ongoing operation. This is advantageously exploited within the framework of the method proposed here, not only to determine a deviation, but also to extrapolate it into the future and thus create a forecast.In one embodiment, the standard operating behavior is defined by the nominal model. The deviation variable thus specifically refers to a deviation between the operating behavior described by the detailed model and the operating behavior of the at least one component described by the nominal model. In one embodiment, the deviation variable is determined by comparing the detailed model with the nominal model. In one configuration, the deviation variable is determined by comparing the second mapping with the first mapping. In one embodiment, the deviation variable is determined repeatedly, in particular periodically; in particular, the comparison of the detailed model with the nominal model is carried out repeatedly, preferably periodically. Alternatively or additionally, the deviation variable is determined in an event-controlled manner; in particular, the comparison of the detailed model with the nominal model is carried out in an event-controlled manner.“Event-driven” is understood to mean that the determination of the deviation variable is initiated or triggered by a triggering event or signal. Such a triggering event can be, for example, the reaching of a predetermined operating time threshold or mileage threshold of the power device or of the at least one component, or the occurrence of a predetermined detected malfunction or the like. In the context of the present technical teaching, an output variable is generally understood to mean any variable that can be determined, in particular calculated, by a control device of the power device - in particular by means of a control or regulating method, for example by means of a model-based predictive method. Accordingly, an input variable is generally understood to mean any variable that can be used to determine, in particular to calculate, an output variable.The input variable can be supplied externally, or alternatively, it can also be calculated. In one embodiment, the input variable can itself be an output variable determined by the control device or received from another control device. It is possible for an output variable to be determined, in particular calculated, as a function of a plurality of input variables. Alternatively or additionally, a plurality of output variables can be determined, in particular calculated, from the at least one input variable. In one embodiment of the method, a value of the at least one input variable used to control the power device is determined by means of a model-based predictive method. In one embodiment, the value of the input variable used to control the power device is calculated using the detailed model.In addition, the detailed model—and also the nominal model—are used as an observer structure, which is also used to observe the operation of the power device and, in particular, to detect deviations. Alternatively, the value of the at least one input variable used to control the power device is determined using another control or regulation method, in particular a characteristic map-based method. In this embodiment, the detailed model and the nominal model are used as an observer structure, which is not used directly to control the power device, but rather—in this case, in particular, exclusively—to observe its operation and, in particular, to detect deviations.In one embodiment, the at least one input variable is also calculated based on the detailed model for the purpose of detecting deviations—in parallel or in addition to determining the value used for control by means of the other control or regulation method. In the context of the present technical teaching, a model-based predictive method is understood in particular to mean model predictive control (MPC). In the context of the present technical teaching, a nominal model is generally understood to mean a model that is formed, in particular, data-enabled, based on existing data, wherein the nominal model is static or fixed during the runtime of the control device, i.e., is not changed or adapted during the runtime of the control device.In the context of the present technical teaching, a detailed model is generally understood to mean an assignment of the at least one output variable to the at least one input variable, which assignment is created based on measured values. In particular, in a simple embodiment, the detailed model can comprise a collection, table, or database of values ​​recorded during operation of the power device for the at least one input variable and the at least one output variable. In another embodiment, the detailed model is a model that can also be used for the control or regulating method - and is also used according to one embodiment. In one embodiment, the detailed model is based on the nominal model, wherein the detailed model, starting from the nominal model, is modified during the runtime of the control device, in particular expanded by measured values ​​obtained at the power device and thus adapted to the operation of the specific power device - preferably in the field of application.In the context of the present technical teaching, inventory data is generally understood to mean data taken from an existing database, preferably obtained before commissioning of the specific control device or power device. In one embodiment, the inventory data is not collected in real time, in particular not during the runtime of the specific control device or power device. The inventory data can, for example, be measured on a test bench or obtained from – particularly highly accurate – models or simulations, or even from the past from the application field of other control devices or power devices.In one embodiment, the nominal model is data-based in such a way that, by varying the at least one input variable, it minimizes a mean square error of the at least one output variable and, at the same time, can extrapolate reliably and meaningfully outside the measurement space included in the data-based calculation. In one embodiment, the detailed model is also data-based in an analogous manner, wherein it is preferably configured to minimize the same error as a function of the at least one input variable; however, the detailed model is expanded during operation, particularly at locations with a high information content, so that the model quality increases. In one embodiment, the output variable is also determined, in particular calculated, based on the detailed model, preferably based on the model-based predictive method, in particular by means of model-predictive control.According to a development of the invention, it is provided that a future deviation in the operating behavior of the at least one component is determined based on the deviation function. In this way, a reliable forecast of the operating behavior of the at least one component can advantageously be obtained. In one embodiment, the future deviation in the operating behavior of the at least one component is determined based on the deviation function for at least one forecast time. It is possible for the future deviation to be determined for a plurality of forecast times. This allows a particularly detailed forecast of the future operating behavior. In one embodiment, future aging of the at least one component is determined based on the deviation function. Alternatively or additionally, future wear of the at least one component is determined based on the deviation function.This approach advantageously allows for reliable planning of needs-based maintenance and / or replacement of the at least one component. According to a further development of the invention, a predicted end of service life for the at least one component is determined based on the deviation function, taking into account a tolerance range predetermined for the at least one component. In one embodiment, the predicted end of service life is determined as the point in time at which, according to the prediction determined based on the deviation function, the predetermined tolerance range is first exceeded or a limit of the predetermined tolerance range is reached.Advantageously, the determination of a predicted end of service life allows for reliable planning of a needs-based replacement of the at least one component and thus avoids, in particular, premature and therefore cost-inefficient replacement as well as unnecessary downtimes of the power device. According to a further development of the invention, the deviation function is extrapolated over time. This represents a procedure for projecting the deviation variable into the future that is both simple and reliable. According to a further development of the invention, a control specification for controlling at least one actuator as the at least one component of the power device is used as the at least one input variable. The control specification can be a direct control variable for an actuator, i.e., a variable that is used directly to control the actuator.Alternatively, the control specification can be a variable as a function of which a further control specification, for example a direct control variable for an actuator, is determined, in particular calculated. In one embodiment, the control specification is a fuel mass, wherein as a function of the fuel mass, at least one direct control variable for suitably controlling a fuel injector or fuel valve can be determined, with which the fuel injector or fuel valve is controlled in order to introduce the fuel mass into a combustion chamber or an air path of the power device. In one embodiment, the control specification is a fuel mass to be introduced into a combustion chamber or an air path of an internal combustion engine. The at least one actuator can be an actuator or actuator of the power device.In particular, the actuator can be an actuator of an engine block of an internal combustion engine, for example an injector, a valve, or a flap. However, the actuator can also be an actuator outside the power device, for example outside an engine block, for example an actuator provided to influence an externally provided cooling circuit, such as a valve, a pump, or the like, or an actuator of a transmission or an electrical device to which the power device is operatively connected. Alternatively or additionally, it is provided that a physical observable detected by at least one sensor as the at least one component of the power device is used as the at least one output variable.In the context of the present technical teaching, a physical observable is generally understood to be a measured variable or measurable variable, for example a temperature, in particular air or exhaust gas temperature, a mass, in particular air mass or exhaust gas mass, a chemical concentration or a partial pressure, a pressure, an electric current, or an electric voltage. In one embodiment, the output variable is an exhaust gas temperature, in particular an exhaust gas temperature of an internal combustion engine. In one embodiment, the input variable as a control specification is a fuel mass to be introduced into a combustion chamber or air path of an internal combustion engine, and the output variable is an exhaust gas temperature of the internal combustion engine.Optionally, additional model variables are implemented in the nominal model and the detailed model as input variables and / or output variables, for example, nitrogen oxide emissions, a power output, in particular engine power output, a peak combustion pressure, a boost pressure, an air mass, a start of injection, a rail pressure, a charge air temperature, and / or a high-temperature cooling circuit temperature. According to a further development of the invention, a plurality of input variables are used as the at least one input variable as an input vector. In one embodiment, the input variables are selected from a group consisting of fuel mass, boost pressure, air mass, start of injection, rail pressure, charge air temperature, and high-temperature cooling circuit temperature. In one embodiment, a plurality of control specifications for controlling an associated actuator are used as a control vector for a plurality of components of the power device.In this case, the control vector is the input vector, and the control specifications are the input variables. Alternatively or additionally, a plurality of output variables is used as the at least one output variable. The output variables are in particular selected from a group consisting of exhaust gas temperature, nitrogen oxide emissions, power, in particular engine power, peak combustion pressure, boost pressure, air mass, start of injection, rail pressure, charge air temperature, and high-temperature cooling circuit temperature. Alternatively or additionally, the at least one deviation variable is determined as a plurality of deviation variables, in particular as a deviation vector. The deviation vector comprises the plurality of deviation variables as vector components. In this way, a plurality of deviations can advantageously be determined in a resource-saving manner.According to a further development of the invention, it is provided that the at least one deviation variable is determined as a variable selected from a group consisting of an input deviation variable and an output deviation variable. In one embodiment, at least two variables are determined as the at least one deviation variable: at least one input deviation variable and at least one output deviation variable. In one embodiment, an input deviation variable is understood to mean a deviation assigned to a control specification. This can be a deviation in the implementation of the control specification by an actuator - for example, a defective, dirty, aged, or worn - wherein the actuator is unable to correctly implement the control specification, for example due to a deficit such as a defect, contamination, wear, or aging.In particular, the input deviation variable is an actuator deviation variable. In one embodiment, an output deviation variable is understood to be a deviation assigned to an observable. This can be a measurement error or sensor error, in particular a systematic one. For example, it can be a measurement error that occurs due to a sensor deficiency, such as a defect, contamination, wear, or aging. According to a further development of the invention, at least one input deviation variable and at least one output deviation variable are determined as the at least one deviation variable, wherein a multi-dimensional temporal development of the at least one input deviation variable and the at least one output deviation variable is determined as the deviation function.In one embodiment, the deviation function is used to infer a future development of a complex component as the at least one component of the power device. In the context of the present technical teaching, a multidimensional temporal development is generally understood to mean a vector-valued deviation function as a function of time. The vector-valued deviation function can be used to infer the future development of the complex component. A complex component is generally understood to be a component that cannot be described by a single control specification and / or a single measured value, whereby the behavior of the complex component is rather determined by a plurality of parameters.A complex component is, for example, an SCR catalyst, the operating behavior of which can be characterized, for example, by a maximum conversion, wherein the maximum conversion in turn is determined by a plurality of measured values ​​and / or control specifications. According to a development of the invention, it is provided that at least one input variable is assigned a detailed output variable based on the detailed model and a nominal output variable based on the nominal model, wherein the at least one deviation variable is calculated based on the at least one detailed output variable and the at least one nominal output variable. In one embodiment, a plurality of input variables are each assigned a detailed output variable and a nominal output variable, in particular in order to assign a detailed output vector and a nominal output vector to an input vector.In particular, a deviation of the behavior of the at least one component from the standard operating behavior can advantageously be determined from a comparison of the detailed output variable with the nominal output variable. In particular, the detailed model is compared with the nominal model by comparing the at least one detailed output variable with the at least one nominal output variable. In one embodiment, starting from at least one specific input variable, a correction variable or a correction vector is calculated which minimizes a deviation between the detailed output variable assigned to the specific input variable according to the detailed model and the nominal output variable assigned to the specific input variable to which the correction variable is applied according to the nominal model. In particular, the correction variable is obtained as the deviation variable, or the deviation variable is determined from the correction variable.According to a further development of the invention, a first Gaussian process model is used as the nominal model. Gaussian process models are particularly suitable for model-predictive control or regulation of a power device: Compared to polynomial-based models, they are particularly easy to adapt to new or changed data points in the application field, and they exhibit more suitable and physically correct behavior in the boundary regions of the given parameter space. Compared to physical models, they require significantly less computational effort. Furthermore, they enable the direct use of test bench data, data from—particularly high-precision—simulations or models, or even field data from the application field.The nominal model as a Gaussian process model is given in particular by stored data points obtained, for example, in test bench experiments or from simulations or models, where XN ^. x m in particular n input variables for m different operating states and with Y N ^ ℝ m x k in particular, k output variables are specified for the m different operating states. Furthermore, the Gaussian process model is characterized by a predetermined calculation scheme for an expected value E ^ ℝ l x k and a variance Var for input variables not included in the original data set for l different operating states X u ^ ℝ n x l given: with a mean function m, a predetermined variance ^ ^ 2 , the identity matrix I, and a covariance function K, which depends on the Euclidean distance r between two points x1, x2 in the following way: with a predetermined distance parameter l and a predetermined signal variance ^^ ^^ . Thus, in equations (1) and (2) ^^ ( ^^ ^^ , ^^ ^^ ) ^ ℝ l x m , ^^ ( ^^ ^^ , ^^ ^^ ) ^ ℝ m x m , I ^ ℝ m x m and YN ^ ℝ m x k . The mean function m is preferably obtained as a Gaussian process model. To obtain the nominal model, a first Gaussian process model, also referred to as the basic grid, is first fitted to second data under at least one constraint derived from first data, i.e., test bench data or data obtained from – in particular, highly accurate – models or simulations. In particular, input variables X N selected, and the corresponding output variables Y Nare calculated in such a way that a deviation of the expected value E of the first Gaussian process model, which is determined from the input variables XN and the output variables YN, from the second inventory data is minimized while observing the constraint. Furthermore, for the purpose of determining the first Gaussian process model, it is preferably assumed that its mean value function m = 0. The first inventory data comprise a larger parameter space than the second inventory data. In particular, it is possible for the first inventory data to be measured on a single-cylinder test bench, while the second inventory data are measured on the full engine or likewise on the single-cylinder test bench and, in the latter case, are preferably converted to the full engine using a simulation model. The constraint is preferably obtained as a trend, for example determining whether certain parameters behave linearly or monotonously with respect to one another.If no such trend is detected, the constraint can be omitted, in which case the adaptation of the first Gaussian process model to the second inventory data is then also referred to as unrestricted. Finally, the nominal model is obtained as the adapted first Gaussian process model. According to a development of the invention, a second Gaussian process model is used as the detailed model. In one embodiment, the second Gaussian process model has the first Gaussian process model as a mean function. The detailed model therefore has the nominal model as a mean function. In particular, the obtained expected value of the first Gaussian process model is used as the mean function m in the detailed model, which now includes the second inventory data as known input variables XD2 and output variables YD2. The input variables X known from the inventory data are thus D2 and output variables Y D2to subsets of the input variables XD and the output variables YD of the detailed model. In one embodiment, the detailed model is adapted according to equations (1) to (4) during operation of the power device, i.e., in the application field. For this purpose, measured values, i.e., newly measured data points, are added, in particular during operation of the power device, or data points from the existing data identified as being capable of improvement are replaced by newly measured data points. In one embodiment, the at least one deviation variable, in particular the deviation vector, is determined in the following way: The detailed model vectors YD of the output variables are calculated in all q detailed model points with respect to the associated input vectors X D summarized: ^^^^, ^^ ^^ ^^ ^^= diag Analogously, all corresponding nominal model vectors Y Nthe output variables in all q detail model points with respect to the corresponding input vectors XD taking into account an input correction vector ^X: ^^^^, ^^ ^^ ^^ ^^(∆ ^^) = diag As an alternative to the additive offset assumed in equation (6), a multiplicative deviation can also be assumed by multiplicatively linking the input correction vector ^X – with a suitable definition – with the input variables XD. From equations (5) and (6), the following optimization problem for the input correction vector ^X can now be obtained: By finding the input correction vector ^X that minimizes the expression in the norm of equation (7), the desired deviation vector—that is, the deviation quantity—is found as the input correction vector ^X that minimizes the expression in the norm of equation (7). This represents the deviation of the detailed model trained using measured values ​​from the nominal model obtained from existing data and thus any deviations in the operating behavior from the standard operating behavior. The described procedure can be extended by considering an output correction vector ^Y as an alternative or in addition to considering the input correction vector ^X in order to detect errors in the output quantities: ^^^^, ^^ ^^ ^^ ^^ ( ∆ ^^ ) = diag For the combined consideration of input correction vector ^X and output correction vector ^Y, the following optimization problem can be solved: [∆ ^^,∆ ^^ ] = arg In particular, the input correction vector ^X and the output correction vector ^Y are optimized separately according to equation (9). In particular, at least one of the two correction vectors ^X, ^Y is now determined as the at least one deviation variable at a first time k and at least one further time k+ ^k; this is formally explained here for the combination of both correction vectors [ ^X, ^Y]k, whereby the application to only one correction vector is trivial. A model for the temporal progression of the at least one deviation variable is now set up, whereby in the simplest case it is assumed that this – in particular starting from the standard operating behavior ( ^X = 0, ^Y = 0) – increases linearly with time: [∆ ^^,∆ ^^] ^^+∆ ^^ = [ ^^ ^^, ^^ ^^] ∙ ∆ ^^ , (10) with the gradients [ ^^ ^^ , ^^ ^^ ], which can be determined, for example, by scalar, i.e., element-wise, recursive least-squares estimation (RLS). For a component such as an actuator or a sensor, a tolerance range is assumed, the limit of which is defined by a maximum permissible deviation [ ^X max , ^Y max ] is given. The predicted end of runtime of an actuator, for example, is then a time FX at which the maximum permissible deviation ^X max is achieved: Accordingly, the predicted end of runtime of a sensor is a time FY at which the maximum permissible deviation is achieved: More complex predictions for complex components such as an SCR catalyst can be achieved by moving the nominal model ^^ ^^, ^^ ( ^^; ^^ ) = ^^ ^^ ( ^^+ ^^ ^^ ^^ ) + ^^ ^^ ^^(13). For example, aging of an SCR catalyst shifts a point of maximum conversion. The end of the SCR catalyst's operating life F is reached when the maximum conversion falls below a certain minimum. This can be formulated for the respective model output in particular as a multi-objective optimization problem: [~, ^^] = argm[ ^^,i ^n^]^ ^^,− ^^ ^^, ^^( ^^; ^^)൧s. t. ^^ ^^, ^^( ^^; ^^) = ^^ ^^ ^^ ^^ . (14)where ^^ ^^ ^^ ^^the lower limit for the maximum conversion as the model output. Similarly, depending on the model output and the physical framework, similar optimization problems can be set up for determining runtime ends or wear points. The object is also achieved by providing a control device for a power device that is configured to carry out a method according to the invention or a method according to one or more of the previously described embodiments. In connection with the control device, in particular, those advantages arise that were already explained above in connection with the method. The control device is configured, in particular, to operate the power device.In one embodiment, the control device is configured to operate an internal combustion engine, an internal combustion engine-generator combination device, a fuel cell, an energy storage device, in particular a battery, an electrolyzer, a data center or microgrid, or another controllable or regulatable load on an electrical network. In one embodiment, the control device is configured to determine the at least one input variable for controlling, in particular for regulating, the power device. In particular, the control device is configured to control the power device with the at least one input variable. Finally, the object is also achieved by creating a power arrangement comprising a power device and a control device according to the invention or a control device according to one or more of the previously described embodiments.In connection with the power arrangement, in particular those advantages arise which were already explained above in connection with the method or the control device. In the context of the present technical teaching, a power device is generally understood to be a device which is configured to provide power, in particular electrical and / or mechanical power, or to convert or consume power. The power device can thus be designed in particular as a power provision device or as a power conversion device. A power provision device is generally understood to be a device which provides power, in particular electrical and / or mechanical power, using electrical, mechanical, chemical or electrochemical energy - or another form of energy.A power conversion device is generally understood to be a device that uses or consumes power, in particular electrical or mechanical power, to convert or store energy, for example, to provide chemical energy in the form of certain substances such as hydrogen or methanol, or electrochemical energy, using electrical energy. In particular, the power device can be an internal combustion engine, a combined internal combustion engine-generator device, i.e., a genset, a fuel cell, an energy storage device, in particular a battery, or an electrolyzer. However, the power device can also be a larger, complex system, for example, comprising a plurality of the aforementioned devices, or in particular also a data center or a microgrid.In particular, the power device can also be a controllable or regulatable load on an electrical network. According to a further development of the invention, the power device is designed as an internal combustion engine, an internal combustion engine-generator combination device, a fuel cell, an energy storage device, in particular a battery, an electrolyzer, a data center or microgrid, or another controllable or regulatable load on an electrical network. The invention is explained in more detail below with reference to the drawing.1 shows a schematic representation of an exemplary embodiment of a power arrangement with an exemplary embodiment of a power device and an exemplary embodiment of a control device; Figure 2 shows a schematic representation of an embodiment of a method for operating the power device; and Figure 3 shows a schematic representation of the functioning of the method according to Figure 2 in the form of a diagram. Figure 1 shows a schematic representation of an exemplary embodiment of a power arrangement 1 with an exemplary embodiment of a power device 3 and an exemplary embodiment of a control device 5. The power device 3 is designed here as an internal combustion engine 4. Alternatively, the power device 3 can be an internal combustion engine-generator compound device or a fuel cell.However, it can also be designed, in particular, as an energy storage device, in particular a battery, electrolyzer, data center, or microgrid, or as another controllable or regulatable load on an electrical network. The control device 5 is operatively connected to the power device 3 in order to control the power device 3. The control device 5 is configured to carry out a method for operating the power device 3, as described below.In this case, at least one deviation variable is determined based on a nominal model formed from existing data, which comprises a first mapping between at least one input variable and at least one output variable, and a detailed model created during operation of the control device 5 based on measured values, which comprises a second mapping between the at least one input variable and the at least one output variable. This deviation variable relates to a deviation of an operating behavior of at least one component 6 of the power device from a standard operating behavior defined by the nominal model. A deviation function is determined as the temporal development of the at least one deviation variable, and a future development of the at least one component 6 is inferred based on the deviation function.In the exemplary embodiment shown here, the at least one input variable is a control specification for the control of at least one actuator 7 of the power device 3 as the at least one component 6. The control specification is preferably a fuel mass, wherein at least one direct control variable for the suitable control of a fuel injector 9 of the power device 3 designed as an internal combustion engine 4 is determined as a function of the fuel mass, with which the fuel injector 9 is controlled in order to introduce the fuel mass into a combustion chamber 11. For the sake of better clarity, only one actuator 7, fuel injector 9 and combustion chamber 11 are each identified with the respective reference numeral, wherein the internal combustion engine 4 can have a plurality of combustion chambers 11, each with an associated actuator 7, namely fuel injector 9.Furthermore, in the exemplary embodiment presented here, the at least one output variable is a physical observable of the power device, in particular an exhaust gas temperature of the internal combustion engine 4. Optionally, additional model variables are implemented in the nominal model and the detailed model as input variables and / or output variables, for example a boost pressure, an air mass, a start of injection, a rail pressure, nitrogen oxide emissions, a power output, in particular an engine power output, a peak combustion pressure, a charge air temperature, and a high-temperature cooling circuit temperature. Fig. 2 shows a schematic representation of an embodiment of a method for operating the power device 3. Identical and functionally identical elements are provided with the same reference numerals in all figures, so that reference is made to the preceding description in each case.In a first step S1, a detailed output variable is assigned to the at least one input variable based on the detailed model. In a second step S2, a nominal output variable is assigned to the at least one input variable based on the nominal model. In a third step S3, the at least one deviation variable is calculated based on the at least one detailed output variable and the at least one nominal output variable, in particular by optimizing an input correction vector and / or an output correction vector, in particular according to equation (7) or equation (9). In a fourth step S4, the deviation function is extrapolated over time, and in a fifth step S5, future aging or future wear of the at least one component 6 is determined based on the time-extrapolated deviation function.Preferably, in the fifth step S5, a predicted end of runtime for the at least one component 6 is determined based on the temporally extrapolated deviation function, taking into account a tolerance range predetermined for the at least one component 6. A first Gaussian process model is preferably used as the nominal model. Preferably, a second Gaussian process model is used as the detailed model. In one embodiment, the detailed model has the nominal model as a mean value function. Preferably, a plurality of input variables is used as an input vector as the at least one input variable. Alternatively or additionally, a plurality of output variables is used as an output vector as the at least one output variable. Alternatively or additionally, the at least one deviation variable is determined as a plurality of deviation variables, in particular as a deviation vector.Preferably, an input deviation variable and / or an output deviation variable is determined as the at least one deviation variable. In a preferred embodiment, at least one input deviation variable and at least one output deviation variable are determined as the at least one deviation variable, wherein a multidimensional temporal development of the at least one input deviation variable and the at least one output deviation variable is determined as the deviation function. Based on the deviation function, conclusions are preferably drawn about a future development of a complex component 6, such as an SCR catalyst. Fig. 3 shows a schematic representation of the functioning of the method according to Fig. 2 in the form of a diagram.In particular, the functioning of the method is explained with reference to Figure 3 using a simplified, low-dimensional representation: Here, an input variable X, which is, for example, a fuel mass, is plotted against an output variable Y, which is, for example, an exhaust gas temperature. It is assumed that one of the fuel injectors 9 of the internal combustion engine 4 exhibits a deviation from its standard operating behavior, in particular a leak, so that when controlled with a certain fuel mass, it actually injects a larger fuel mass. The black, filled crosses represent the relationship between the input variable X and the output variable Y according to the detailed model; the open circles represent the relationship between the input variable X and the output variable Y according to the nominal model.Due to the deviation in the form of the leakage of the fuel injector 9, the detailed model differs from the nominal model representing the standard operating behavior in that the fuel masses used to control the fuel injectors 9 are systematically assigned exhaust gas temperatures that are too high, since in reality more fuel is always injected than corresponds to the fuel mass used as the control specification.At a specific sampling time, starting from a specific fuel mass Xa, an additive or multiplicative correction vector ^X is calculated, for which a deviation is minimized between an exhaust gas temperature Ya,D(Xa) calculated as a detailed output variable according to the detailed model as a function of the fuel mass Xa and an exhaust gas temperature Ya,N(Xr = Xa + ^X) calculated from the nominal model as a nominal output variable as a function of a fuel mass Xr to which the correction vector ^X is applied – or in the multiplicative case Ya,N(Xr = Xa ∙ ^X). In this way, with the thus optimized correction vector ^X, a deviation value is obtained that allows a conclusion to be drawn about the existing deviation from the standard operating behavior, in this case about the leakage of the fuel injector 9.Alternatively or in addition to determining an (input) correction vector ^X for the input variable, it is possible to determine an output correction vector ^Y for the output variable in an analogous manner and thus, for example, to detect and preferably correct a sensor error. The detailed model points represent the current situation of the power device 3 at the sampling time, while the nominal model points represent an average deviation-free case for the same input situation. In the multidimensional case, the actual deviation can be deduced from the obtained multidimensional input correction vector ^X and / or output correction vector ^Y, particularly by means of artificial intelligence or a neural network. The deviation can be verified for plausibility using additional input and / or output variables.For example, it can be decided that simultaneously occurring deviations in boost pressure and air mass of the internal combustion engine 4 indicate a clogged air filter, or an injector fault can be verified by comparing exhaust gas temperature and nitrogen oxide emissions. This procedure is then repeated for a plurality of sampling times, from which different values ​​are obtained for the at least one deviation variable at the different sampling times, thus a discrete deviation function as the temporal development of the at least one deviation variable. This allows a temporal extrapolation of the deviation function, preferably in the form of a linear regression.In a simple manner, a predicted end of service life for at least one component 6, for example the fuel injector 9, can then be deduced from a gradient determined for the deviation function, taking into account a predetermined tolerance range.

Claims

CLAIMS 1. Method for operating a power device (3) by means of a control device (5), wherein ˗ based on a nominal model formed from inventory data, which comprises a first mapping between at least one input variable and at least one output variable, and a detailed model created during operation of the control device (5) on the basis of measured values, which detailed model comprises a second mapping between the at least one input variable and the at least one output variable, at least one deviation variable is determined, which relates to a deviation of an operating behavior of at least one component (6) of the power device (3) from a standard operating behavior, wherein ˗ a deviation function is determined as a temporal development of the at least one deviation variable, wherein ˗ a future development of the at least one component (6) is inferred based on the deviation function. 2.Method according to claim 1, wherein a future deviation in the operating behavior, in particular future aging or future wear of the at least one component (6) is determined based on the deviation function.

3. Method according to one of the preceding claims, wherein a predicted end of runtime for the at least one component (6) is determined based on the deviation function, taking into account a tolerance range predetermined for the at least one component (6).

4. Method according to one of the preceding claims, wherein the deviation function is extrapolated over time.

5. Method according to one of the preceding claims, wherein ˗ a control specification for controlling at least one actuator (7) as the at least one component (6) of the power device (3) is used as the at least one input variable, and / or wherein. ˗ a physical observable detected by at least one sensor as the at least one component (6) of the power device (3) is used as the at least one output variable.

6. Method according to one of the preceding claims, wherein ˗ a plurality of input variables is used as the at least one input variable as an input vector, in particular a plurality of control specifications for controlling a respective associated actuator (7) as a control vector of a plurality of components (6) of the power device (3), and / or wherein ˗ a plurality of output variables is used as the at least one output variable as an output vector, and / or wherein ˗ the at least one deviation variable is determined as a plurality of deviation variables, in particular as a deviation vector. 7.Method according to one of the preceding claims, wherein the at least one deviation variable is determined to be a variable selected from a group consisting of an input deviation variable and an output deviation variable.

8. Method according to one of the preceding claims, wherein the at least one deviation variable is determined to be at least one input deviation variable and at least one output deviation variable, wherein the deviation function is determined to be a multidimensional temporal development of the at least one input deviation variable and the at least one output deviation variable, and wherein, optionally, the deviation function is used to infer a future development of a complex component as the at least one component (6) of the power device (3). 9.Method according to one of the preceding claims, wherein at least one input variable of the at least one input variable is assigned a detailed output variable based on the detailed model and a nominal output variable based on the nominal model, and wherein the at least one deviation variable is calculated based on the at least one detailed output variable and the at least one nominal output variable.

10. The method according to one of the preceding claims, wherein a first Gaussian process model is used as the nominal model.

11. The method according to claim 10, wherein a second Gaussian process model is used as the detailed model, which in particular has the first Gaussian process model as a mean value function.

12. A control device (5) for a power device (3), wherein the control device (5) is configured to carry out a method according to one of the preceding claims.

13. A power arrangement (1) comprising a power device (3) and a control device (5) according to claim 12. 14.Power arrangement (1) according to claim 13, wherein the power device (3) is designed as an internal combustion engine (4), an internal combustion engine-generator combination device, a fuel cell, an energy storage device, in particular a battery, an electrolyzer, a data center or microgrid, or another controllable or regulatable load on an electrical network.

Images